A single-cell big data analysis system and method

Abstract

A single cell big data analysis system, comprising a high-throughput sequencing data acquisition module, a single cell public data module, a batch correction module, an intelligent annotation module, a mass statistics module, a personalized analysis module and a visualization output module, characterized in that the high-throughput sequencing data acquisition module downloads the reference genome FASTA file of the required species from the Ens emb l, UCSC or NCBI public database.The single cell big data analysis system and method have the advantages that a large number of public databases are covered, the clustering annotation work is completed by using the latest algorithm, various accurate intelligent annotation results are given, mass statistics is completed for each annotation result, the analysis efficiency is improved, a visualization and process-based domestic single cell data analysis platform is established, the drawing and selection of intelligent charts are realized, the re-mining of CNS level research results is realized, and the advantages that the research and development investment of scientific researchers is reduced and the efficient and reasonable utilization of data is realized are achieved.

Description

Technical Field

[0001] This invention belongs to the field of intelligent technology for high-throughput sequencing data analysis, specifically, it relates to a system and method for single-cell big data analysis. Background Technology

[0002] Single-cell RNA sequencing (scRNA-seq) technology can independently analyze each cell in a single sample, for example, up to ten thousand cells, generating massive, high-dimensional datasets. According to literature, hundreds of algorithms and workflows exist for processing this data. However, selecting the appropriate algorithm has become one of the most critical challenges in current single-cell data analysis workflows. Furthermore, single-cell data analysis is highly dependent on programming environments such as R or Python. While these two languages ​​each have their advantages in statistical computing and machine learning, data exchange and integration between them is not easy.

[0003] Therefore, ensuring seamless integration between different algorithms and analysis tools while maintaining efficiency and accuracy, and achieving comprehensive analysis of single-cell data has become a pressing technical problem in this field. Summary of the Invention

[0004] The purpose of this invention is to propose a single-cell big data analysis system. The system and method for single-cell big data analysis are described below.

[0005] Another objective of this invention is to propose a single-cell big data analysis method.

[0006] To achieve the above objectives, the technical solution of the present invention is:

[0007] A single-cell big data analysis system includes a high-throughput sequencing data acquisition module, a single-cell public data module, a batch correction module, an intelligent annotation module, a massive statistics module, a personalized analysis module, and a visualization output module, characterized in that:

[0008] The high-throughput sequencing data acquisition module downloads the reference genome FASTA file of the required species from public databases such as Ensembl, UCSC, or NCBI.

[0009] The single-cell public data module includes publicly available single-cell data from dozens of species, including humans, mice, rats, pigs, rhesus monkeys, Père David's deer, fruit flies, silkworms, zebrafish, salamanders, rice, and Arabidopsis thaliana.

[0010] The batch correction module includes an algorithm flow developed based on MNN, graph-based, and deep-learn, as well as a cross-species analysis flow;

[0011] The intelligent annotation module includes hundreds of organization-specific intelligent annotation models built using the AysenseBio2.0 annotation system.

[0012] The massive statistics module includes annotation statistical plotting, inter-group difference plotting, and complex analysis plotting of annotation X grouping, GSEA, and GSVA analysis.

[0013] The personalized analysis module includes the ability to automatically perform dozens of analyses, such as pseudo-temporal analysis, intercellular communication analysis, deconvolution analysis, and transcription factor analysis.

[0014] The visualization output module includes the output of various interactive analysis charts such as cluster diagrams, bubble charts, heatmaps, and communication difference charts.

[0015] The single-cell big data analysis system of the present invention can also be further implemented using the following technical measures.

[0016] In the aforementioned single-cell big data analysis system, the high-throughput sequencing data acquisition module, matrix generation module, batch correction module, single-cell public data module, intelligent annotation module, massive statistics module, personalized analysis module, and visualization output module are connected in sequence, and the output data of the previous module matches the input data of the next module.

[0017] The aforementioned single-cell big data analysis system includes modules such as high-throughput sequencing data acquisition, matrix generation, batch correction, single-cell public data, intelligent annotation, massive statistics, personalized analysis, and visualization output, all of which can run automatically.

[0018] The aforementioned single-cell big data analysis system, wherein the high-throughput sequencing data acquisition module further includes using the mkfastq command to convert BCL files into FASTQ format files, performing sequence alignment of the FASTQ format files with the corresponding reference genome, and converting them into a count matrix file.

[0019] A single-cell big data analysis method, characterized by the following steps:

[0020] a. High-throughput sequencing data acquisition module, including the following steps:

[0021] (1) Download the reference genome FASTA file of the required species from the Ensembl, UCSC or NCBI public database, along with the corresponding GTF file, and use the mkref command to create a new reference genome database;

[0022] (2) Use the mkfastq command to convert the BCL file to FASTQ format file, and perform data quality control and data filtering on low-quality reads;

[0023] (3) Align the FASTQ format file with the corresponding reference genome; use CellRanger alignment software to read the FASTQ data in each sample catalog, align each read sequence to the reference genome database, and decode the cell barcode and unique feature identification sequence.

[0024] (4) Convert to a matrix file of count; after obtaining the decoded cell barcode, count the decoded cell barcode and the unique feature recognition sequence to construct a gene-cell expression matrix;

[0025] b. A single-cell public database module, including publicly available single-cell data of dozens of species such as humans, mice, pigs, rhesus monkeys, Père David's deer, fruit flies, silkworms, zebrafish, salamanders, rice, and Arabidopsis thaliana. The publicly available single-cell data is uniformly processed into standard RDS format data and standardized cell annotation is performed. The processing method includes the following steps.

[0026] Various matrix data were downloaded and organized from the GEO and Single Cell Portal databases. After undergoing a unified normalization correction algorithm, the UMI numbers assigned to specific cells were summed to obtain the total sequencing depth cell attribute. This cell attribute was then used in a negative binomial error distribution and log-linked function regression model to construct a regression model, correcting for sequencing depth differences between cells and standardizing the data.

[0027] log(E(xi))=β0+β1log10m;

[0028] Where: xi is the UMI count vector of gene i, m is the molecular vector assigned to the cell, i.e., mj = i xi j. The solution to the regression model problem is a set of parameters: intercept β0 and slope β1. The dispersion parameter θ of the underlying distribution is unknown. This parameter is estimated by fitting a negative binomial regression model.

[0029] c. Batch correction module, including algorithmic processes developed based on MNN, graph-based, and deep-learn, as well as cross-species analysis processes:

[0030] (1) Calculate the similarity matrix among all cells in each batch. Let X be an n×p matrix, where each row represents the expression profile data of a cell. The similarity matrix S is defined as:

[0031] S = exp(-l[z;-rllP / 202)

[0032] Where: σ is a parameter that controls the distance decay rate;

[0033] (2) By calculating the similarity matrix among all cells in each batch, the most similar set of anchor points in each batch is determined. Let Sb represent the similarity matrix among cells in the b-th batch, and define a cost function C:

[0034]

[0035] Where: A is a set of |A| anchor points, Ab represents the set of anchor points in the b-th batch, λ is a regularization coefficient used to penalize cases where the number of anchor points is too large, and the goal is to minimize the cost function C;

[0036] Each non-anchor cell is mapped to the low-dimensional space containing its nearest anchor. Let x represent the expression profile data of the i-th cell and ak represent the expression profile of the k-th anchor.

[0037] d. The intelligent annotation module, which uses the AysenseBio 2.0 annotation system to build hundreds of organization-specific intelligent annotation models, includes the following steps:

[0038] (1) Input a scRNA-seq dataset containing known cell types as the training set. Each cell sample has an expression profile, that is, the expression level of each gene.

[0039] (2) Select genes with discriminative power from the training data as features, wherein the genes have significant expression differences between different cell types;

[0040] (3) Suppose there are n cell types, or labels. For each label i, train a regression model, which is expressed as:

[0041]

[0042] Where: y i It is the binary output of the i-th label, which is either 0 or 1;

[0043] x is the input feature vector, representing the gene expression level;

[0044] w i It is the weight vector associated with the i-th class;

[0045] b i It is the bias term, the intercept;

[0046] It is a logic function used to map the result of a linear combination to a probability interval of 0 and 1;

[0047] (4) After establishing the regression model, for a new input matrix x, use the regression model to calculate a probability value for each cell label, return the probability distribution of all labels, and input the relevant parameters again to complete the statistical annotation;

[0048] e. The massive statistics module includes annotated statistical plotting, between-group difference plotting, and complex analysis plotting of annotated X-groups, GSEA, and GSVA analyses, including the following steps:

[0049] (1) After annotating all cell type subpopulations, continue to statistically analyze the differences of all cell groups in multiple groups, including using T-test, multi-group ANOVA, and RO / E algorithm to compare differences between groups.

[0050] (2) For each inter-group difference, gene set variation analysis is calculated. Let f(x) represent the estimated kernel density of gene expression value x within a certain gene set in the sample. The positive enrichment fraction ES+ is calculated using the following formula:

[0051] Where: F(x) is the cumulative distribution function, s represents a specific sample, and for the negative enrichment score ES-, it is a low expression region, and the enrichment score ES is the difference between ES+ and ES-.

[0052] (3) For each annotation result and each group, an expression matrix is ​​formed again, and then analyzed:

[0053] Where: xi and yi are two variables;

[0054] and The expression mean of the disease group and the control group;

[0055] ∑ represents the summation symbol, which sums over all sample points;

[0056] f. Personalized analysis module, including automated completion of dozens of analyses such as pseudo-temporal analysis, inter-cell communication, deconvolution, and transcription factor analysis;

[0057] The personalized statistics module has built a workflow that automatically completes grouped time series analysis, inter-group communication difference analysis, Bulk-seq deconvolution, and subgroup-specific transcription factor analysis, while performing parallel processing to greatly optimize the analysis process.

[0058] g. Visualization output module, including the output of various interactive analysis charts such as cluster diagrams, bubble charts, heatmaps, and communication difference charts. It groups the data through hierarchical clustering algorithm, calculates the differential genes of specific groups, draws bubble charts and heatmaps, selects specific subpopulations for intercellular communication analysis, calculates the differences in communication data according to groups, and finally obtains communication difference images for single-cell big data analysis.

[0059] The single-cell big data analysis method of the present invention can also be further implemented using the following technical measures.

[0060] In the aforementioned method, the parameter for controlling the distance decay rate is set to 0.1 or 0.01.

[0061] In the aforementioned method, the intelligent annotation module outputs dozens of annotation results at once.

[0062] In the aforementioned method, the massive statistics module calculates the inter-group differences for each output result.

[0063] In the aforementioned method, the two variables are the observed values ​​of the disease group and the control group.

[0064] After adopting the above technical solution, the single-cell big data analysis system and method of the present invention have the following advantages:

[0065] 1. It encompasses numerous public databases and uses the latest algorithms to complete clustering and annotation work. It also provides multiple accurate annotation results and performs massive statistics on each intelligent annotation result, turning complex single-cell data analysis results into multiple-choice questions, greatly optimizing the analysis content and improving analysis efficiency.

[0066] 2. A visualized and streamlined domestic single-cell data analysis platform has been established, enabling the creation and selection of intelligent charts, facilitating the re-mining of CNS-level research results, assisting researchers in reducing R&D investment, and achieving efficient and rational use of data. Attached Figure Description

[0067] Figure 1 For the batch correction of this embodiment of the invention, the optimal anchor point is found to form the final clustering diagram;

[0068] Figure 2 This is a schematic diagram illustrating the annotation details obtained by the intelligent annotation system of this invention when annotating cell subpopulations.

[0069] Figure 3 This is a schematic diagram illustrating the comparison of differences between groups using algorithms such as T-test, multi-group ANOVA, and RO / E, in an embodiment of the present invention to show the proportion statistics and functional clustering data between two groups.

[0070] Figure 4 This is a functional analysis diagram of GSVA according to an embodiment of the present invention;

[0071] Figure 5 This is a schematic diagram illustrating the automated completion of dozens of analyses, such as pseudo-temporal analysis, intercellular communication, deconvolution, and transcription factor analysis, in the personalized analysis module of this invention.

[0072] Figure 6 This is a schematic diagram illustrating the output of various interactive analysis diagrams, such as clustering diagrams, bubble diagrams, heatmaps, and communication difference diagrams, in an embodiment of the present invention.

[0073] Figure 7 This is a flowchart of the process for single-cell big data analysis according to the present invention. Detailed Implementation

[0074] The present invention will be further described below with reference to the embodiments and accompanying drawings.

[0075] Example 1

[0076] Please see now Figure 7 , Figure 7 This is a flowchart of the single-cell big data analysis system of the present invention. The single-cell big data analysis system of the present invention includes a high-throughput sequencing data acquisition module, a single-cell public data module, a batch correction module, an intelligent annotation module, a massive statistics module, a personalized analysis module, and a visualization output module.

[0077] The high-throughput sequencing data acquisition module downloads the reference genome FASTA file of the required species from public databases such as Ensembl, UCSC, or NCBI.

[0078] The single-cell public data module includes publicly available single-cell data from dozens of species, including humans, mice, rats, pigs, rhesus monkeys, Père David's deer, fruit flies, silkworms, zebrafish, salamanders, rice, and Arabidopsis thaliana.

[0079] The batch correction module includes an algorithm flow developed based on MNN, graph-based, and deep-learn, as well as a cross-species analysis flow;

[0080] The intelligent annotation module includes hundreds of organization-specific intelligent annotation models built using the AysenseBio2.0 annotation system.

[0081] The massive statistics module includes annotation statistical plotting, inter-group difference plotting, and complex analysis plotting of annotation X grouping, GSEA, and GSVA analysis.

[0082] The personalized analysis module includes the ability to automatically perform dozens of analyses, such as pseudo-temporal analysis, intercellular communication analysis, deconvolution analysis, and transcription factor analysis.

[0083] The visualization output module includes the output of various interactive analysis charts such as cluster diagrams, bubble charts, heatmaps, and communication difference charts.

[0084] The high-throughput sequencing data acquisition module, matrix generation module, batch correction module, single-cell public data module, intelligent annotation module, massive statistics module, personalized analysis module, and visualization output module are connected in sequence, with the output data of the previous module matching the input data of the next module.

[0085] The high-throughput sequencing data acquisition module, matrix generation module, batch correction module, single-cell public data module, intelligent annotation module, massive statistics module, personalized analysis module, and visualization output module can all run automatically.

[0086] The high-throughput sequencing data acquisition module also includes using the mkfastq command to convert BCL files into FASTQ format files, performing sequence alignment of the FASTQ format files with the corresponding reference genome, and converting them into a count matrix file.

[0087] Example 2

[0088] Please see now Figure 1-6 The single-cell big data analysis method of the present invention includes the following steps:

[0089] a. High-throughput sequencing data acquisition module, including reference genome database construction, data quality control, data filtering, sequence alignment, quantitative analysis, and matrix generation, including the following steps:

[0090] (1) Download the reference genome FASTA file of the required species and the corresponding GTF (Gene Transfer Format) file from the Ensembl, UCSC or NCBI public database, and use the mkref command to create a new reference genome database;

[0091] (2) Use the mkfastq command to convert the BCL file to a FASTQ file, and perform data quality control and data filtering on low-quality reads;

[0092] (3) Perform sequence alignment between the FASTQ format file and the corresponding reference genome;

[0093] (4) Convert to a matrix file with count as the main element; Figure 1 In this embodiment of the invention, after batch correction, the optimal anchor point is found to form the final clustering diagram.

[0094] b. A single-cell public database module includes publicly available single-cell data from dozens of species, including humans, mice, rats, pigs, rhesus monkeys, Père David's deer, fruit flies, silkworms, zebrafish, salamanders, rice, and Arabidopsis thaliana. Due to the inconsistent data formats in publicly available literature, the data needs to be uniformly processed into standard RDS format and annotated with standardized cell annotations. This method includes the following steps;

[0095] Various matrix data were downloaded and organized from databases such as GEO and Single Cell Portal, and then normalized and corrected using a unified algorithm, specifically as follows:

[0096] The summation of all UMI (unique molecular identifiers) numbers assigned to a specific cell is used as the total sequencing depth attribute. This cell attribute is then used in a negative binomial error distribution and log link function regression model to construct a regression model that corrects for sequencing depth differences between cells and standardizes the data.

[0097] log(E(xi))=β0+β1log10m

[0098] Where xi is the UMI count vector of gene i, and m is the molecular vector assigned to the cell, i.e., mj = i xi j. The solution to this regression problem is a set of parameters: intercept β0 and slope β1. The dispersion parameter θ of the underlying NB distribution is also unknown and needs to be estimated from the data, using the NB parameterization form with mean μ and variance μ + μ2θ;

[0099] c. The batch correction module includes algorithmic processes developed based on MNN (Mutual Nearest Neighbors), graph-based methods, and deep-learning methods, as well as cross-species analysis processes.

[0100] 1. Calculate the similarity matrix among all cells in each batch. Assume X is an n×p matrix, where each row represents the expression profile data of one cell. The similarity matrix S can be defined as:

[0101] S ij =exp(-||x i -x j || 2 / 2σ 2 )

[0102] Where σ is a parameter that controls the distance decay rate, and it is usually taken to be a relatively small value, such as 0.1 or 0.01.

[0103] 2. Next, we need to determine the most similar set of anchor points in each batch. This can be done by calculating the similarity matrix between all cells in each batch. Let Sb represent the similarity matrix between cells in the b-th batch, then we can define a cost function C:

[0104] Where A is a set of |A| anchor points, Ab represents the set of anchor points in the b-th batch, and λ is a regularization coefficient used to penalize cases where the number of anchor points is too large. The goal is to minimize the cost function C.

[0105] Each non-anchor cell is mapped to the low-dimensional space containing its nearest anchor. Specifically, let xi represent the expression profile data of the i-th cell, and ak represent the expression profile of the k-th anchor.

[0106] d. The intelligent annotation module includes hundreds of organization-specific intelligent annotation models and the self-developed AysenseBio2.0 annotation system. Figure 2 This is a schematic diagram illustrating the annotation details obtained by the annotation system of this invention for annotating cell subpopulations.

[0107] 4. Input a scRNA-seq dataset containing known cell types as the training set. Each cell (sample) has its expression profile, i.e., the expression level of each gene.

[0108] 5. Select discriminative genes as features from the training data. These genes show significant expression differences between different cell types.

[0109] 6. Assuming there are n cell types (or labels), then for each label i, train a regression model. This model can be represented as:

[0110]

[0111] ·y i It is the binary output (0 or 1) of the i-th label class.

[0112] x is the input feature vector (e.g., gene expression level).

[0113] ·w i It is the weight vector associated with the i-th class.

[0114] ·b i It is the bias term (intercept).

[0115] · It is a sigmoid function used to map the result of a linear combination to a probability interval of (0,1).

[0116] 4. After establishing the model, for a new input matrix x, use the model to calculate a probability value for each cell label, return the probability distribution of all labels, and input the relevant parameters again to complete the statistical annotation.

[0117] e. The massive statistics module includes annotated statistical plots, inter-group difference plots, and complex analysis plots for annotated X-groups, as well as GSEA and GSVA analyses.

[0118] 1. After the annotation results of the subgroups, the statistical data is further divided into multiple groups, including using T-tests, multi-group ANOVA, and RO / E algorithms to compare differences between groups. Most importantly, the annotation module outputs dozens of annotation results at once, and the big data statistics module calculates the differences between groups for each output result.

[0119] 2. For each type of inter-group difference, Gene Set Variation Analysis is calculated. If f(x) represents the kernel density estimate of gene expression value x within a gene set in the sample, then the positive enrichment score ES+ is calculated as follows:

[0120]

[0121] Where F(x) is the cumulative distribution function, and s represents a specific sample. For a negative enrichment score ES-, low-expression regions are considered. The final enrichment score ES may be the difference between ES+ and ES-.

[0122] 3. For each annotation result and each group, an expression matrix will be generated again, and relevant analysis will be performed:

[0123] • Here, xi and yi are two variables, such as the observed values ​​of the disease group and the control group.

[0124] · and The expression mean of the disease group and the control group.

[0125] • ∑ represents the summation symbol, which means summing over all sample points.

[0126] f. Personalized analysis module, including automated completion of dozens of analyses such as pseudo-time series, inter-cell communication, deconvolution, and transcription factor analysis.

[0127] The personalized statistics module has already built a workflow that can automatically complete tasks such as grouped time series analysis, inter-group communication difference analysis, Bulk-seq deconvolution, and subgroup-specific transcription factor analysis, while also considering parallel processing to significantly optimize the analysis process.

[0128] g. Visualization output module, including various interactive analysis charts such as cluster diagrams, bubble charts, heatmaps, and communication difference charts. Data is grouped using a hierarchical clustering algorithm, and differentially expressed genes within specific groups are calculated to generate bubble charts and heatmaps. Specific subpopulations are then selected for intercellular communication analysis; the communication data is further analyzed by grouping and differences are calculated to ultimately obtain communication difference images.

[0129] This invention possesses substantial features and significant technological advancements. Its single-cell big data analysis system and method encompass numerous public databases and utilize state-of-the-art algorithms for clustering and annotation, providing multiple precise and intelligent annotation results. Furthermore, it performs massive statistical analysis on each annotation result. These steps transform complex single-cell data analysis into a multiple-choice process, significantly optimizing the analysis content and improving efficiency. A visualized and streamlined domestic single-cell data analysis platform has also been established, enabling the creation and selection of intelligent charts and facilitating the re-mining of CNS-level research findings. Through these efforts, researchers can reduce R&D investment and achieve efficient and rational data utilization.

[0130] The single-cell big data analysis system and method of this invention includes a high-throughput sequencing data acquisition module, a matrix generation module, a batch correction module, a single-cell public data module, an intelligent annotation module, a massive statistics module, a personalized analysis module, and a visualization output module. These modules are connected in the order of analysis; the output data of the previous module matches the input data of the next module. The workflow of multiple modules can run automatically. Testing showed that the analysis of 500,000 single-cell data can be completed in just two days. This invention utilizes a modular operation scheme to establish a visualized and streamlined domestic single-cell data analysis system, greatly reducing the difficulty of single-cell data analysis and improving the analysis efficiency of researchers.

[0131] The above embodiments are for illustrative purposes only and are not intended to limit the invention. Those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention. Therefore, all equivalent technical solutions should also fall within the scope of this invention and are defined by the claims.

Claims

1. A single-cell big data analysis system, comprising a high-throughput sequencing data acquisition module, a single-cell public data module, a batch correction module, an intelligent annotation module, a massive statistics module, a personalized analysis module, and a visualization output module, characterized in that: The high-throughput sequencing data acquisition module downloads the reference genome FASTA file of the required species from public databases such as Ensembl, UCSC, or NCBI. The single-cell public data module contains publicly available single-cell data from humans, mice, rats, pigs, rhesus monkeys, Père David's deer, fruit flies, silkworms, zebrafish, salamanders, rice, and Arabidopsis thaliana. The batch correction module includes an algorithm flow developed based on MNN, graph-based, and deep-learn, as well as a cross-species analysis flow; The intelligent annotation module includes hundreds of organization-specific intelligent annotation models built using the AysenseBio2.0 annotation system. The massive statistics module includes annotation statistical plotting, inter-group difference plotting, and complex analysis plotting of annotation X grouping, GSEA, and GSVA analysis. The personalized analysis module includes the ability to automatically perform dozens of analyses, such as pseudo-temporal analysis, intercellular communication analysis, deconvolution analysis, and transcription factor analysis. The visualization output module includes the output of various interactive analysis charts such as cluster diagrams, bubble charts, heatmaps, and communication difference charts; a) High-throughput sequencing data acquisition module, including the following steps: (1) Download the reference genome FASTA file of the required species from the Ensembl, UCSC or NCBI public database, along with the corresponding GTF file, and use the mkref command to create a new reference genome database; (2) Use the mkfastq command to convert the BCL file to FASTQ format file, and perform data quality control and data filtering on low-quality reads; (3) Align the FASTQ format file with the corresponding reference genome; use CellRanger alignment software to read the FASTQ data in each sample catalog, align each read sequence to the reference genome database, and decode the cell barcode and unique feature identification sequence; (4) Convert to a matrix file of count; after obtaining the decoded cell barcode, count the decoded cell barcode and the unique feature recognition sequence to construct a gene-cell expression matrix; b. A single-cell public database module, including publicly available single-cell data from humans, mice, rats, pigs, rhesus monkeys, Père David's deer, fruit flies, silkworms, zebrafish, salamanders, rice, and Arabidopsis thaliana. The publicly available single-cell data is uniformly processed into standard RDS format data and standardized cell annotation is performed. The processing method includes the following steps. Various matrix data were downloaded and organized from the GEO and Single Cell Portal databases. After undergoing a unified normalization correction algorithm, the UMI numbers assigned to specific cells were summed to obtain the total sequencing depth cell attribute. This cell attribute was then used in a negative binomial error distribution and log-linked function regression model to construct a regression model, correcting for sequencing depth differences between cells and standardizing the data. log(E(xi)) = β0 + β1 log10 m; Where: xi is the UMI count vector of gene i, m is the molecular vector assigned to the cell, i.e., mj = i xi j, the solution to the regression model problem is a set of parameters: intercept β0 and slope β1, the dispersion parameter θ of the underlying distribution is unknown, and this parameter is estimated by fitting a negative binomial regression model. c. Batch correction module, including algorithmic processes developed based on MNN, graph-based, and deep-learn, as well as cross-species analysis processes: (1) Calculate the similarity matrix among all cells in each batch. Let X be an n×p matrix, where each row represents the expression profile data of a cell. The similarity matrix S is defined as: in: σ It is a parameter that controls the rate of distance decay; (2) By calculating the similarity matrix among all cells in each batch, determine the most similar set of anchor points in each batch, let... Sb Let b be the similarity matrix between cells in the b-th batch, and define a cost function. C : Where: A is a set consisting of |A| anchor points. Ab Indicates the first b Batch of anchor point set, λ It is a regularization coefficient used to penalize cases with an excessive number of anchor points, with the goal of minimizing the cost function. C ; Map each non-anchor cell to the low-dimensional space containing its nearest anchor cell, and let... x Indicates the first i Expression profile data of individual cells ak Indicates the first k The expression spectrum of each anchor point; d. The intelligent annotation module, which uses the AysenseBio 2.0 annotation system to build hundreds of organization-specific intelligent annotation models, includes the following steps: (1) Input a scRNA-seq dataset containing known cell types as the training set. Each cell sample has an expression profile, i.e., the expression level of each gene. (2) Select genes with discriminative power from the training data as features, wherein the genes have significant expression differences between different cell types; (3) Suppose there are n cell types, or labels. For each label i, train a regression model, which is expressed as: in: It is the binary output of the i-th label, which is either 0 or 1; x It is the input feature vector, representing the gene expression level; Is with the first i Class-related weight vectors; It is the bias term, the intercept; It is a logic function used to map the result of a linear combination to a probability interval of 0 and 1.

2. The single-cell big data analysis system as described in claim 1, characterized in that, The high-throughput sequencing data acquisition module, matrix generation module, batch correction module, single-cell public data module, intelligent annotation module, massive statistics module, personalized analysis module, and visualization output module are connected in sequence, with the output data of the previous module matching the input data of the next module.

3. The single-cell big data analysis system as described in claim 1, characterized in that, The high-throughput sequencing data acquisition module, matrix generation module, batch correction module, single-cell public data module, intelligent annotation module, massive statistics module, personalized analysis module, and visualization output module can all run automatically.

4. The single-cell big data analysis system as described in claim 1, characterized in that, The high-throughput sequencing data acquisition module also includes using the mkfastq command to convert BCL files into FASTQ format files, performing sequence alignment of the FASTQ format files with the corresponding reference genome, and converting them into a count matrix file.

5. A single-cell big data analysis method applied to the system of claim 1, characterized in that... Includes the following steps: a. High-throughput sequencing data acquisition module, including the following steps: (1) Download the reference genome FASTA file of the required species from the Ensembl, UCSC or NCBI public database, along with the corresponding GTF file, and use the mkref command to create a new reference genome database; (2) Use the mkfastq command to convert the BCL file to FASTQ format file, and perform data quality control and data filtering on low-quality reads; (3) Align the FASTQ format file with the corresponding reference genome; use CellRanger alignment software to read the FASTQ data in each sample catalog, align each read sequence to the reference genome database, and decode the cell barcode and unique feature identification sequence; (4) Convert to a matrix file of count; after obtaining the decoded cell barcode, count the decoded cell barcode and the unique feature recognition sequence to construct a gene-cell expression matrix; b. The single-cell public database module contains publicly available single-cell data from humans, mice, rats, pigs, rhesus monkeys, Père David's deer, fruit flies, silkworms, zebrafish, salamanders, rice, and Arabidopsis thaliana. The publicly available single-cell data is uniformly processed into standard RDS format data and standardized cell annotation is performed. The processing method includes the following steps. Various matrix data were downloaded and organized from the GEO and Single Cell Portal databases. After undergoing a unified normalization correction algorithm, the UMI numbers assigned to specific cells were summed to obtain the total sequencing depth cell attribute. This cell attribute was then used in a negative binomial error distribution and log-linked function regression model to construct a regression model, correcting for sequencing depth differences between cells and standardizing the data. log(E(xi)) = β0 + β1 log10 m; Where: xi is the UMI count vector of gene i, m is the molecular vector assigned to the cell, i.e., mj = i xi j, the solution to the regression model problem is a set of parameters: intercept β0 and slope β1, the dispersion parameter θ of the underlying distribution is unknown, and this parameter is estimated by fitting a negative binomial regression model. c. Batch correction module, including algorithmic processes developed based on MNN, graph-based, and deep-learn, as well as cross-species analysis processes: (1) Calculate the similarity matrix among all cells in each batch. Let X be an n×p matrix, where each row represents the expression profile data of a cell. The similarity matrix S is defined as: in: σ It is a parameter that controls the rate of distance decay; (2) By calculating the similarity matrix among all cells in each batch, determine the most similar set of anchor points in each batch, let... Sb Let b be the similarity matrix between cells in the b-th batch, and define a cost function. C : Where: A is a set consisting of |A| anchor points. Ab Indicates the first b Batch of anchor point set, λ It is a regularization coefficient used to penalize cases with an excessive number of anchor points, with the goal of minimizing the cost function. C ; Map each non-anchor cell to the low-dimensional space containing its nearest anchor cell, and let... x Indicates the first i Expression profile data of individual cells ak Indicates the first k The expression spectrum of each anchor point; d. The intelligent annotation module, which uses the AysenseBio 2.0 annotation system to build hundreds of organization-specific intelligent annotation models, includes the following steps: (1) Input a scRNA-seq dataset containing known cell types as the training set. Each cell sample has an expression profile, i.e., the expression level of each gene. (2) Select genes with discriminative power from the training data as features, wherein the genes have significant expression differences between different cell types; (3) Suppose there are n cell types, or labels. For each label i, train a regression model, which is expressed as: in: It is the binary output of the i-th label, which is either 0 or 1; x It is the input feature vector, representing the gene expression level; Is with the first i Class-related weight vectors; It is the bias term, the intercept; It is a logic function used to map the result of a linear combination to a probability interval of 0 and 1; (4) After establishing the regression model, for a new input matrix x, use the regression model to calculate a probability value for each cell label, return the probability distribution of all labels, and input the relevant parameters again to complete the statistical annotation; e. The massive statistics module includes annotated statistical plotting, between-group difference plotting, and complex analysis plotting of annotated X-groups, GSEA, and GSVA analyses, including the following steps: (1) After annotating all cell type subpopulations, continue to statistically analyze the differences of all cell groups in multiple groups, including using T-test, multi-group ANOVA, and RO / E algorithm to compare differences between groups; (2) For each group difference, gene set variation analysis is calculated. Let f(x) represent the estimated kernel density of gene expression value x in a certain gene set in the sample. The positive enrichment fraction ES+ is calculated using the following formula: Where: F(x) is the cumulative distribution function, s represents a specific sample, and for the negative enrichment score ES-, it is a low expression region, and the enrichment score ES is the difference between ES+ and ES-. (3) For each annotation result and each group, an expression matrix is ​​formed again, and analyzed: in: xi and yi These are two variables; The expression mean of the disease group and the control group; ∑ represents the summation symbol, which sums over all sample points; f. Personalized analysis module, including automated completion of dozens of analyses such as pseudo-temporal analysis, inter-cell communication, deconvolution, and transcription factor analysis; The personalized statistics module has built a workflow that automatically completes grouped time series analysis, inter-group communication difference analysis, Bulk-seq deconvolution, and subgroup-specific transcription factor analysis, while performing parallel processing to greatly optimize the analysis process. g. Visualization output module, including the output of various interactive analysis charts such as cluster diagrams, bubble charts, heatmaps, and communication difference charts. It groups the data through hierarchical clustering algorithm, calculates the differential genes of specific groups, draws bubble charts and heatmaps, selects specific subpopulations for intercellular communication analysis, calculates the differences in communication data according to groups, and finally obtains communication difference images for single-cell big data analysis.

6. The single-cell big data analysis method as described in claim 5, characterized in that, The parameter for controlling the distance decay rate is set to 0.1 or 0.

01.

7. The single-cell big data analysis method as described in claim 5, characterized in that, The intelligent annotation module outputs dozens of annotation results at once.

8. The single-cell big data analysis method as described in claim 5, characterized in that, The massive statistics module calculates the inter-group differences for each output result.

9. The single-cell big data analysis method as described in claim 5, characterized in that, The two variables are the observed values ​​of the disease group and the control group.

Images